Autonomous Manufacturing of Composite Parts by a Multi-Robot System

Abstract

Aerospace structures require a combination of low weight and high mechanical performance and thus often involve composite materials, e. g. carbon fiber reinforced plastics (CFRP) or fiber metal laminates (FML). The laminate structure involves complex layups, which makes manual production error prone. Today, there still is a lack of innovative production techniques to achieve competitive production rates. Automating this processes demands a smart and flexible, however robust system directly linked to the CAX-chain. We investigated a combination of cooperation industrial robots and computer vision without teaching of the robots. Commercial products like Delmia or Process Simulate are useful for digital factory planning. Add-ons like Cenit’s Fast Suite extend the functionality towards digital production, but lack the ability of handling huge numbers of cut-pieces. It was shown in previous work that a robot’s target points for gripping and dropping cut-pieces can be derived automatically and subsequently the layup can be carried out autonomously. Also was shown that computer vision strongly improves process accuracy and robustness. The focus of this paper lies on the practical implementation of a smart manufacturing execution system in a multi-robot environment. Major components of the cyber-physical system are the Manufacturing Execution System (MES), the robots and their controllers, one ore multiple computer vison systems for detection of the goods being handled, and a simulation environment called CoCo for collision avoidance. This work considers pick-and-place processes, which consist of the steps picking, transfer, dropping and post-drop treatment. Our scenario was an airplane skin demonstrator made of dry CFRP sheets. The jig for placing the cut-pieces is half-shell shaped with a diameter of approx. 4 m and a length of approx. 2 m, while the 108 cut-pieces are approx. 1.2 m by 1.8 m and approx. 1.2 m by 0.8 m in size and are provided in a drawer based storage system. For determining where to grip we use one computer-vision system mounted to a KUKA Quantec KR210 R3100 robot. A second, identical robot with identical grippers is operated using the image coordinates of the same camera. Both robots are mounted to one linear axis of 8 m length. The information on which cut-piece to grip, where to grip it from, what is it’s contour and where to exactly move the grippers for gripping and dropping is contained in a proprietary Job Definition File format (jdf), which can be considered a preliminary step to later extensions of the CPS by a production planning and execution agent. A parser converts the generic jdf-information to an action list for each robot comprising setting the tool center point, move the robot, switch the grippers and do the post-drop tacking. The robots receive their actions by the KUKA technology package Ethernet KRL. Time critical movements, like cooperative cut-piece transfer with two robots, are first parametrized and then executed synchronously by the KUKA technology package RoboTeam. Thus, the fully automated, autonomous production of a generic airplane part with multiple robots in a complex environment could be demonstrated.